The authors introduce ISC, which computes step importance through targeted parameter-level interventions.
Interventional step credit prunes redundant chain-of-thought while preserving accuracy
The authors score each step's causal contribution to the answer and use it to rescale reinforcement learning rewards, cutting tokens by up to 62%.
Top University
Guanda Li · Xuefeng Bai · Yulong Chen · Zhiyong Wang · Kehai Chen · Min Zhang
Harbin Institute of Technology, Shenzhen · University of Aberdeen · University of Cambridge
Research Digest··2 min read
Guanda Li and colleagues propose interventional step credit (ISC), a causal measure of how much each chain-of-thought step influences the final answer, and use it to train models via interventional guidance policy optimization (IGPO).
§